Intra-individual variability in many behavioral and medical constructs, including pain, sleep, cognitive performance, and emotion is of growing interest. However, intraindividual variability is often quantified using the intra-individual standard deviation (iSD), or related measures like variance or the coefficient of variability, all of which are statistically inefficient and unreliable when calculated on relatively few observations. One underused methodological tool for modeling variance is mixed-effects location scale (MELS) modeling. Unlike typical multi-level models, MELS allows for random effects on variance components, alongside random slopes and intercepts, which provides the ability to ask questions about inter-individual differences in intra-individual variability. The present tutorial will introduce the MELS model with an application to data on chronic pain symptoms over time. In this tutorial, we teach the MELS model and demonstrate software through a series of three increasingly sophisticated applications of MELS modeling to pain data. We will begin with a basic MELS model that only estimates variances and unique intercepts for individuals; this will give an opportunity to describe the MELS model, including how the inter- and intra-individual variance components are related to those in multilevel modeling. Next, we will discuss including random slopes, as we anticipate many researchers may need to detrend their data. Finally, we will discuss a MELS model with predictor(s) to account for differences in inter- and intra-individual variability.
Whitaker et al. (Fri,) studied this question.